Return AdjFloat of an AdjFloat summation (#181)

* Return AdjFloat of an AdjFloat summation and check the control update in rf

pyadjoint/adjfloat.py

@@ -342,7 +342,8 @@ def __init__(self, *args):
             self.add_dependency(dep)
 
     def recompute_component(self, inputs, block_variable, idx, prepared):
-        return self.operator(*(term.saved_output for term in self.terms))
+        output = self.operator(*(term.saved_output for term in self.terms))
+        return self._outputs[0].saved_output._ad_convert_type(output)
 
     def __str__(self):
         return f"{self.terms[0]} {self.symbol} {self.terms[1]}"

pyadjoint/reduced_functional.py

@@ -2,6 +2,7 @@
 from .enlisting import Enlist
 from .tape import get_working_tape, stop_annotating, no_annotations
 from .overloaded_type import OverloadedType, create_overloaded_object
+from .adjfloat import AdjFloat
 
 
 def _get_extract_derivative_components(derivative_components):
@@ -196,6 +197,21 @@ def __call__(self, values):
         if len(values) != len(self.controls):
             raise ValueError("values should be a list of same length as controls.")
 
+        for i, value in enumerate(values):
+            control_type = type(self.controls[i].control)
+            if isinstance(value, (int, float)) and control_type is AdjFloat:
+                value = self.controls[i].control._ad_convert_type(value)
+            elif not isinstance(value, control_type):
+                if len(values) == 1:
+                    raise TypeError(
+                        "Control value must be an `OverloadedType` object with the same "
+                        f"type as the control, which is {control_type}"
+                    )
+                else:
+                    raise TypeError(
+                        f"The control at index {i} must be an `OverloadedType` object "
+                        f"with the same type as the control, which is {control_type}"
+                    )      
         # Call callback.
         self.eval_cb_pre(self.controls.delist(values))
 

pyadjoint/tape.py

@@ -29,7 +29,7 @@ def continue_annotation():
 
 class set_working_tape(ContextDecorator):
     """Set a new tape as the working tape.
-    
+
     This class can be used in three ways:
        1) as a free function to replace the working tape,
        2) as a context manager within which a new tape is set as the working tape,

tests/firedrake_adjoint/test_burgers_newton.py

@@ -95,14 +95,15 @@ def J(ic, solve_type, timestep, steps, V):
         solver = NonlinearVariationalSolver(problem)
 
     tape = get_working_tape()
+    J = 0.0
     for _ in tape.timestepper(range(steps)):
         if solve_type == "NLVS":
             solver.solve()
         else:
             solve(F == 0, u, bc)
         u_.assign(u)
-
-    return assemble(u_*u_*dx + ic*ic*dx)
+        J += assemble(u_*u_*dx + ic*ic*dx)
+    return J
 
 
 @pytest.mark.parametrize("solve_type, checkpointing",

tests/firedrake_adjoint/test_external_modification.py

@@ -14,7 +14,7 @@ def test_external_modification():
     v1 = Function(fs)
     v2 = Function(fs)
 
-    u.assign(1.)
+    u.interpolate(1.)
     v1.project(u)
     with stop_annotating(modifies=u):
         u.dat.data[:] = 2.
@@ -23,4 +23,4 @@ def test_external_modification():
     J = assemble(v1*dx + v2*dx)
     Jhat = ReducedFunctional(J, Control(u))
 
-    assert np.allclose(J, Jhat(2))
+    assert np.allclose(J, Jhat(Function(fs).interpolate(2.)))

tests/firedrake_adjoint/test_tlm.py

@@ -66,7 +66,7 @@ def test_tlm_bc():
     c.block_variable.tlm_value = Function(R, val=1)
     tape.evaluate_tlm()
 
-    assert (taylor_test(Jhat, Constant(c), Constant(1), dJdm=J.block_variable.tlm_value) > 1.9)
+    assert (taylor_test(Jhat, c, Function(R, val=1), dJdm=J.block_variable.tlm_value) > 1.9)
 
 
 def test_tlm_func():
@@ -234,7 +234,7 @@ def test_projection():
 
     k.block_variable.tlm_value = Constant(1)
     tape.evaluate_tlm()
-    assert(taylor_test(Jhat, Constant(k), Constant(1), dJdm=J.block_variable.tlm_value) > 1.9)
+    assert(taylor_test(Jhat, k, Function(R, val=1), dJdm=J.block_variable.tlm_value) > 1.9)
 
 
 def test_projection_function():